Can Face ID ID locos and cars ?

[up1950s]

I thought occurred to me on that firebox thread in page 13 where the RFID was brought up . Can Apple Face ID ID a loaded image of many items . and can that be applied to ID a train which would be displayed as a readout assigned to it ? If so small camera heads could be hidden to read the train and cars as they pass , maybe assign routes for it . It could be us and we could be a data entry clerk   oh joy .

[Jbub]

Possibly, and most likely not as complex as facial id. All it would need to do is see the number and identify that way.

[C855B]

Not using face recognition systems - those depend on a baseline system that IDs noses, eyes, etc., and then measures the distances between. It doesn't necessarily compare raw images, although that could be a second level of verification in some systems.

What Richie is talking about could either be OCR or image comparison. I designed and built an OCR data collection system in the '90s, and the basic concepts for it would frankly be easily adaptable to this idea. I would favor image comparison since you're capturing the image anyway and OCR adds a processing layer, and code libraries for image analysis are more plentiful.

[DKS]

Not using face recognition systems - those depend on a baseline system that IDs noses, eyes, etc., and then measures the distances between. It doesn't necessarily compare raw images, although that could be a second level of verification in some systems.

What Richie is talking about could either be OCR or image comparison. I designed and built an OCR data collection system in the '90s, and the basic concepts for it would frankly be easily adaptable to this idea. I would favor image comparison since you're capturing the image anyway and OCR adds a processing layer, and code libraries for image analysis are more plentiful.

How would the system differentiate between cars that, aside from road numbers (which could be very small), are all identical?

[learmoia]

Oddly enough, I have heard that at least 1 Railroad tech company is developing 'Facial ID' style AI technology to use cameras to replace AEI Tag readers.. Where they read the car and pick out initial and number.

~Ian 

[DKS]

More and more I'm thinking RFID is more viable, especially now that DIY readers are in the $10-12 range. But I'm not sure why I care--I won't ever be building a layout big enough to warrant it.

[C855B]

How would the system differentiate between cars that, aside from road numbers (which could be very small), are all identical?

Three things come to mind. First is a layered approach, where 99% of the IDs would be based on image comparisons, after which any ambiguities would be passed to OCR.

Second is a matter of resolution for the OCR. And it doesn't take a quantum leap in tech. An interesting factoid from my particular system's development is that it absolutely did not work with 300ppi scanners. 400ppi? Bingo, 99.4%* accuracy. We're definitely at a point in imaging technology that could easily resolve N scale road numbers. Be thinking license plate readers (which I was working on 12 years ago, but we were scooped by another company and dropped our program).

* - Problem I had was the company standard was >99.7%, based on manual entry + manual verification. Never quite got there. We used the system anyway. At that time we couldn't know that <80% accuracy was perfectly fine for the internet.   

Third is in a system with a predictable ambiguity, you focus on what it takes to resolve. It wouldn't be necessary to evaluate the entire image, in the loading of the baseline you would highlight whatever area you expect to be the differentiation. I would expect that at database loading time, the software would flag a new image as too close to something already in the file, "please show me where to find the difference."

IOW, it's doable, with reasonable effort. ROI as a MRR product? Yeah. Right.

[DKS]

ROI as a MRR product? Yeah. Right.

Thought so.

[railnerd]

I thought occurred to me on that firebox thread in page 13 where the RFID was brought up . Can Apple Face ID ID a loaded image of many items . and can that be applied to ID a train which would be displayed as a readout assigned to it ? If so small camera heads could be hidden to read the train and cars as they pass , maybe assign routes for it . It could be us and we could be a data entry clerk   oh joy .

It is all possible— there are application programming interfaces for machine vision in both iOS and Android. NFC/RFID is probably easier, though.

My friend Seth Neumann has an RFID reader on his yard throat so as he pulls a train in from staging, destinations get assigned and a switch list is printed.

-Dave

[railnerd]

It is all possible— there are application programming interfaces for machine vision in both iOS and Android. NFC/RFID is probably easier, though.

My friend Seth Neumann has an RFID reader on his yard throat so as he pulls a train in from staging, destinations get assigned and a switch list is printed.

-Dave

and like DKS, since I'm working on a small N-Scale switching layout, I don't have an interest at home.

The HO club I am in already has more than 2000 pieces of rolling stock, so we think the "changeover" to RFID would also be super painful.  We had a whole program to convert the fleet to metal wheels that took many years— now we need to go back and add resistors for block detection— and when we do that, we may add the 125kHz (or 13.56MHz) tags to the trucks.

If only each member had taken home 25-50 cars we could have made better use of the g-d pandemic shutdown.

-Dave

[ncbqguy]

Dave
Plant this idea with your RFID guys....
RFID says, “Hi, I’m 4521 and I am HERE.   With that you don’t need a separate block detection system.   Just readers wherever a detection is needed.   The signal logic can be a subset of the RFID program, filtering out readers that aren’t needed for signals and ignoring any input not associated with a train.  One (who is clever with computers) could even parse out direction of movement with sequence of detection to feed the signal logic  and “OS” trains going by stations, even tying it into voice sequences “OS 4521 West by Cookville 22:15” or whatever the correct protocol should be....and it would squawk at at the dispatcher.
Charlie Vlk

[DKS]

Dave
Plant this idea with your RFID guys....
RFID says, “Hi, I’m 4521 and I am HERE.   With that you don’t need a separate block detection system.   Just readers wherever a detection is needed.   The signal logic can be a subset of the RFID program, filtering out readers that aren’t needed for signals and ignoring any input not associated with a train.  One (who is clever with computers) could even parse out direction of movement with sequence of detection to feed the signal logic  and “OS” trains going by stations, even tying it into voice sequences “OS 4521 West by Cookville 22:15” or whatever the correct protocol should be....and it would squawk at at the dispatcher.
Charlie Vlk

Exactly. And now with the proliferation of cheap readers, it wouldn't be cost-prohibitive. Some of those cheap DIY readers include USB interfaces, which would greatly simplify integrating everything into a cohesive system that could even--wait for it--detect breaks in a train.

[railnerd]

My friend Don Marenzi IS using RFID on his locos to help his dispatcher— whenever an engine passes a station, the dispatcher can see it.  His layout is TT&TO so this is a big labor saving device which cuts down on the number of "agent operators" that he needs to run.

We're all a bit stir crazy out here in the SF Bay Area because we've been under strict local rules for a long time— so our weekly in person "bull sessions" at lunch on Fridays haven't happened since March.  Eventually, I can find out from him how it is working out, but other than a few large clubs (like my own club Silicon Valley Lines) that have the ability to do social distancing, nobody has been hosting operating sessions.

At SVL, we have built some crazy "rolling web cam" cars to allow for remote operation.  Since each flatcar has a webcam, we have thought about running a QR code reader tasks to call out DTC blocks as trains enter and leave.

-Dave

[daniel_leavitt2000]

Several years ago we had a detective canvas the department showing us all a very grainy video of a suspect vehicle (i think it was a hit and run bu I can not be sure). All that was really visible was the basic sedan outline, and tail light position. Even the shape of the tail lights was blury.

I had this idea - make an app that can scan a photo or video and tell you what the make/model/year of the vehicle in question would be. I asked a software engineer that was a member of the local train club about it. It COULD be done, but would cost a lot of money. This was before smartphones has an AI stack built into the SOC so I imagine it would be a bit easier to implement today.

Face ID, at least for Apple and MS are NOT able to do this for a very specific reason. The Face ID modules are actually physically separated from the rest of the device hardware. This system has its own sandbox, boot and detect sequences that make any data they contain inaccessible to the rest of the system. This is done for Touch ID as well. You can see the interface for these on Windows Device Manager under Trusted Platform Module. Keep in mind that this is just the driver to pull data from the TPM, and does not grant access to its own firmware.

I believe that TPM hardware and whatever Apple uses actually boots before BIOS or equivalent so that even root kits can not access it.

All this means is that those ID cameras are not accessible in any way to the phone hardware.